AI Planning in Context AI Planning in the

Скачать презентацию AI Planning in Context AI Planning in the

4869ca1097c10ea54d6a8f4d1ee96323.ppt

Количество слайдов: 38

AI Planning in Context AI Planning in the Context of Domain Modelling, Task Assignment and Execution

Overview • • • Context of Practical Systems Context of Task Assignment & Execution Context of Multiple Agents Context of Plan Representation & Use Example Practical Planners Planning++

Edinburgh AI Planners in Productive Use

Practical AI Planners Planner Reference Applications STRIPS Fikes & Nilsson 1971 Mobile Robot Control, etc. HACKER Sussman 1973 Simple Program Generation NOAH Sacerdoti 1977 Mechanical Engineers Apprentice Supervision NONLIN Tate 1977 Electricity Turbine Overhaul, etc. NASL Mc. Dermott 1978 Electronic Circuit Design OPM Hayes-Roth & Hayes-Roth 1979 Journey Planning ISIS-II Fox et. al. 1981 Job Shop Scheduling (Turbine Production) MOLGEN Stefik 1981 Experiment Planning in Molecular Genetics DEVISER Vere 1983 Spacecraft Mission Planning FORBIN Miller et al. 1985 Factory Control SIPE/SIPE-2 Wilkins 1988 Crisis Action Planning, Oil Spill Management, etc. SHOP/SHOP-2 Nau et al. 1999 Evacuation Planning, Forest Fires, Bridge Baron, etc. I-X/I-Plan Tate et al. 2000 Emergency Response, etc.

Course Reading • • • Review of AI Planners to 1990 Hendler, J. A. , Tate, A. and Drummond, M. (1990) “AI Planning: Systems and Techniques”, AI Magazine Vol. 11, No. 2, pp. 61 -77, Summer 1990, AAAI Press. http: //aaaipress. org/ojs/index. php/aimagazine/article/download/833/751 Knowledge-Based Planners Wilkins, D. E. and des. Jardins, M. (2001) “A Call for Knowledge-based Planning”, AI Magazine, Vol. 22, No. 1, pp. 99 -115, Spring 2001, AAAI Press. http: //www. aaai. org/ojs/index. php/aimagazine/article/view/1547/ or http: //www. ai. sri. com/pub_list/808 • • O-Plan Paper Tate, A. and Dalton, J. (2003) “O-Plan: a Common Lisp Planning Web Service”, invited paper, in Proceedings of the International Lisp Conference 2003, October 12 -25, 2003, New York, NY, USA, October 12 -15, 2003. http: //www. aiai. ed. ac. uk/project/ix/documents/2003 -luc-tate-oplan-web. pdf • • Optimum-AIV Paper Tate, A. (1996) “Responsive Planning and Scheduling Using AI Planning Techniques – Optimum-AIV”, in “Trends & Controversies – AI Planning Systems in the Real World”, IEEE Expert: Intelligent Systems & their Applications, Vol. 11 No. 6, pp. 4 -12, December 1996. http: //www. aiai. ed. ac. uk/project/oplan/documents/1996/96 -ieee-is-trends-and-controversies-orig. pdf • • • SHOP/SHOP 2 Applications Paper Nau, D. , Au, T-C. , Ilghami, O. , Kuter, U. , Wu, D. , Yaman, F. , Muñoz-Avila, H. , and Murdock, J. W. (2005) Applications of SHOP and SHOP 2, IEEE Intelligent Systems, March-April 2005, Vol. 20, No. 2, pp. 34 -41, Computer Society. http: //www. cs. utexas. edu/~chiu/papers/Nau 05 shop 2. pdf Other Practical Planners Ghallab, M. , Nau, D. and Traverso, P. (2004) “Automated Planning – Theory and Practice”, Chapters 19, 22 and 23, Elsevier/Morgan Kaufmann.

Origins of some well known AI Planners Hendler, Tate and Drummond, AI Magazine, 1990

Overview • Context of Practical Systems Context of Task Assignment & Execution • Context of Multiple Agents • Context of Plan Representation & Use • Example Practical Planners • Planning++

Dynamic Planning Description of Σ Initial State Objectives Execution Status Planner Plans Controller Observations Actions System Σ Events • problem: real world differs from model described by Σ • more realistic model: interleaved planning and execution – plan supervision – plan revision – re-planning • dynamic planning: closed loop between planner and controller – execution status

O-Plan 3 Levels of Agents: Task Assignment, Planning & Execution Task Assigner Capabilities Planner Constraints Plan State Domain Info Capabilities Executor Constraints Plan State Domain Info Capabilities Constraints Plan State Domain Info

Overview • Context of Practical Systems • Context of Task Assignment & Execution Context of Multiple Agents • Context of Plan Representation & Use • Example Practical Planners • Planning++

Multiple-Agent Platforms behind some Practical AI Planners • Multiple Planning Agents (MPA) platform is the basis for the SRI International SIPE Planner • The Open Planning Architecture is the basis for O-Plan and is designed to handle multiple planner roles and levels, such as task assigner, planning specialists, plan execution • I-X is intended to support multiple types of command, sensemaking, analysis, planning (I-Plan), decision making, execution and communications agents even in mixed agent frameworks.

Multiple Agents in the Context of Communications for Emergency Response Collaboration and Communication Central Authorities Command Control Emergency Responders Isolated Personnel

Example I-X Multiagent Applications

People & Organizations Environment Adapted from H. Kitano and S. Tadokoro, Robo. Cup Rescue A Grand Challenge for Multiagent and Intelligent Systems, AI Magazine, Spring, 2001. (aaai. org) Copyright (c) 2001, Association for the Advancement of Artificial Intelligence Robo. Rescue 50 Year Programme Systems

for Multiagent and Intelligent Systems, AI Magazine, Spring, 2001. Adapted from H. Kitano and S. Tadokoro, Robo. Cup Rescue A Grand Challenge

Overview • Context of Practical Systems • Context of Task Assignment & Execution • Context of Multiple Agents Context of Plan Representation & Use • Example Practical Planners • Planning++

Plan Representation & Use • Plan representation itself an important area. • Plans are used in many areas beyond activity planning … such as situation understanding and summarisation, natural language interpretation and generation, etc. • Plans provide an ontological and formal representation core for a wide range of practical applications and uses.

Uses of a Plan Representation Knowledge Acquisition User Communication Plan Representation Formal Analysis System Manipulation

Plan Representation & Use • AI planning work has influenced standards related to process and plan representations used by many industries and fields. E. g. , • • MIT Process Handbook Process Interchange Format (PIF) NIST Process Specification Language (PSL) DARPA Shared Planning & Activity Representation (SPAR) ISO 18629 Industrial Automation Systems and Integration Process Specification Language

Overview • • Context of Practical Systems Context of Task Assignment & Execution Context of Multiple Agents Context of Plan Representation & Use Example Practical Planners • Planning++

Nonlin (1974 -1977) • • • Hierarchical Task Network Planning Partial Order Planner Plan Space Planner Goal structure-based plan development – only considers alternative “approaches” based on plan rationale QA/“Modal Truth Criterion” condition achievement Condition “types” to limit search Allows for multiple “contributors” to achieve facts for plan robustness “Compute Conditions” for links to external data bases and systems (attached procedures) Operations Research algorithms for time and resource constraints • Nonlin core is a basis for text book descriptions of HTN Planning

O-Plan (1983 -1999) Features • • • Domain knowledge elicitation and modelling tools Rich plan representation and use Hierarchical Task Network Planning Detailed constraint management Goal structure-based plan monitoring Dynamic issue handling Plan repair in low and high tempo situations Interfaces for users with different roles Management of planning and execution workflow

O-Plan Unix Sys Admin Aid

O-Plan MOUT Task Description, Planning and Workflow Aids

Try out O-Plan as an Example Planner • Web accessible HTN AI Planner: See http: //www. aiai. ed. ac. uk/project/oplan/web-demo/ • Try the Unix Systems administration script generator. Consider other applications for which this generation technique may be suitable. Try a few block stacking examples, and ponder why the “Sussman Anomaly” task was not able to be solved by early AI planners. Try the “three pigs” resource constrained house building examples. Look at the domain and task description file. Can you explain why some tasks need little or no search and others more? Why does one task have no solution in the given domain? • • • : am-cycles = Agenda management cycles (problem solving cycles) : n-alts-chosen = Number of alternatives chosen. 0 means the planner had no search at all : n-alts-remaining = Number of alternatives remaining. Indicating choices possible. : n-poisons = Number of dead ends reached (diagnostic - should be same as : n-alts-chosen)

Optimum-AIV

Optimum-AIV (1992 -4) Features • • Based on O-Plan design Rich plan representation and use Hierarchical Task Network (HTN) Planning Detailed constraint management Plan and User rationale recorded Dynamic issue handling Plan repair using test failure recovery plans Integration with ESA’s Artemis Project Management System

Typical Features of Practical AI Planners • • • Hierarchical Task Network (HTN) Planning Partial Order Planning (POP) Rich domain model Detailed constraint management, simulations and analyses Integration with other systems (user interfaces, databases, spreadsheets, project management systems, etc).

Overview • • • Context of Practical Systems Context of Task Assignment & Execution Context of Multiple Agents Context of Plan Representation & Use Example Practical Planners Planning++

Planning Research Areas & Techniques – Domain Modelling – Domain Description – Domain Analysis – – – – – Search Methods Graph Planning Algtms Partial-Order Planning Hierarchical Planning Refinement Planning Opportunistic Search Constraint Satisfaction Optimisation Method Issue/Flaw Handling – Plan Analysis – Plan Simulation – Plan Qualitative Modelling HTN, SIPE PDDL, NIST PSL TIMS Heuristics, A* Graph. Plan Nonlin, UCPOP NOAH, Nonlin, O-Plan Kambhampati OPM CSP, OR, TMMS NN, GA, Ant Colony Opt O-Plan NOAH, Critics Qineti. Q Excalibur – Plan Repair – Re-planning – Plan Monitoring O-Plan, IPEM – Plan Generalisation – Case-Based Planning – Plan Learning Macrops, EBL CHEF, PRODIGY SOAR, PRODIGY – User Interfaces – Plan Advice – Mixed-Initiative Plans SIPE, O-Plan SRI/Myers TRIPS/TRAINS – Planning Web Services O-Plan, SHOP 2 – Plan Sharing & Comms – NL Generation – Dialogue Management I-X, … …

Planning Research Areas & Techniques se n se s – e Generalisation k Plan que Search Methods Heuristics, A* a– Case-Based Planning Graph Planning Algtms Graph. Plan m – hni. Learning Partial-Order Planning Nonlin, UCPOP to ec User Interfaces Hierarchical Planning NOAH, Nonlin, O-Plan – s Refinement Planning Kambhampati i e t – Plan Advice Plans Opportunistic Search OPM m TMMS s – Mixed-Initiative Constraint Satisfaction e OR, h Colony l. CSP, GA, t. Ante Opt – Planning Web Services Optimisation Method b NN, l Issue/Flaw Handling ro O-Plan P f al – Plan Sharing & Comms – NL Generation o – Domain Modelling – Domain Description – Domain Analysis – – – – – Plan Analysis – Plan Simulation – Plan Qualitative Modelling HTN, SIPE PDDL, NIST PSL TIMS NOAH, Critics Qineti. Q Excalibur – Plan Repair – Re-planning – Plan Monitoring – Dialogue Management O-Plan, IPEM Macrops, EBL CHEF, PRODIGY SOAR, PRODIGY SIPE, O-Plan SRI/Myers TRIPS/TRAINS O-Plan, SHOP 2 I-X, … … Deals with whole life cycle of plans

A More Collaborative Planning Framework • Human relatable and presentable objectives, issues, sense-making, advice, multiple options, argumentation, discussions and outline plans for higher levels • Detailed planners, search engines, constraint solvers, analyzers and simulators act in this framework in an understandable way to provide feasibility checks, detailed constraints and guidance • Sharing of processes and information about process products between humans and systems • Current status, context and environment sensitivity • Links between informal/unstructured planning, more structured planning and methods for optimisation

I-X/I-Plan (2000 - ) • Shared, intelligible, easily communicated and extendible conceptual model for objectives, processes, standard operating procedures and plans: – – I N C A Issues Nodes/Activities Constraints Annotations • Communication of dynamic status and presence for agents, and their collaborative processes and process products • Context sensitive presentation of options for action • Intelligent activity planning, execution, monitoring, re-planning and plan repair via I-Plan and I-P 2 (I-X Process Panels)

I-X aim is a Planning, Workflow and Task Messaging “Catch All” • Can take ANY requirement to: – – Handle an issue Perform an activity Respect a constraint Note an annotation • Deals with these via: – – – Manual activity Internal capabilities External capabilities Reroute or delegate to other panels or agents Plan and execute a composite of these capabilities (I-Plan) • Receives reports and interprets them to: – Understand current status of issues, activities and constraints – Understand current world state, especially status of process products – Help user control the situation • Copes with partial knowledge of processes and organisations

Domain Editor I-X for Emergency Response Process Map Tool Panel Messeng I-Plan

Summary • • • Context of Practical Sysetms Context of Task Assignment & Execution Context of Multiple Agents Context of Plan Representation & Use Example Practical Planners Planning++